I have been doing a bit of reading about the new(ish) surface tools in ArcGIS Pro that are delivered under the 'Surface toolset', and in particular the ability to define maximum 'neighbourhood distance' and 'adaptive neighbourhoods' to surface processing.

By way of a little background, my employers are in the process of acquiring a regionwide holding of LiDAR data as part of the national mapping agencies national elevation project (https://www.linz.govt.nz/products-services/data/types-linz-data/elevation-data).

The advent of region wide LiDAR coverage will be a huge benefit to all and it is our intention to mine this dataset as fully and in as robust a manner as possible. With LiDAR data, one of the principal issues we have struggled with is how to derive LiDAR DEM based metrics which adequately account for the variable nature in scale of natural terrain. By this I mean that fine scale features such as small incised streams etc. are freely intermixed with larger features such as hills and valleys. The aim here is to produce meaningful information such as slope groupings that can be applied at a paddock scale.

Traditional matrix based approaches tend to use small windows (3x3, 7x7) to say derive slope and this results in slope information far too detailed and noisy to be practically applied at a paddock scale. Making the window size larger to capture large terrain features by contrast, quickly becomes unwieldy and generalises out important local landscape features.

Manifold have very kindly implemented the excellent sieve tool which goes a huge way to mitigating excess detail in a robust way (and is required to clean up the outputs from the ESRI tools).

It seems however, that on the face of it this new approach from ESRI represents an excellent way to address this problem with LiDAR derivatives and I am wondering if others on the forum have also struggled with multi scale landscape issues?

I am still trying to find out more about this tool, but I would be very interested to hear if anyone has any positive or negative thoughts about such an approach as on the face of it, I would love to see similar implemented in Manifold with their usual panache :)

https://pro.arcgis.com/en/pro-app/latest/tool-reference/spatial-analyst/how-surface-parameters-works.htm#ESRI_SECTION1_837390E28AA843B3AFD882AC3F1BD710

https://www.esri.com/arcgis-blog/products/arcgis-pro/analytics/new-slope-aspect-curvature/

https://pro.arcgis.com/en/pro-app/latest/tool-reference/spatial-analyst/surface-parameters.htm

Landsystems Ltd ... Know your land | www.landsystems.co.nz

This reminds me the wavelet approach of surface analysis in order to get rid of "noise"

https://infoscience.epfl.ch/record/142935

Thanks for this apo. There is some reading there, though the abstract sounds like the research aims to address exactly this issue I am describing. I will continue to read as time permits as this is very pertinent to GIS in resource management.

Landsystems Ltd ... Know your land | www.landsystems.co.nz

https://www.esri.com/arcgis-blog/products/arcgis-pro/analytics/new-slope-aspect-curvature/

Based on the high-level description in the above link, we already have around half of what the new ArcGIS feature is adding, eg, (a) we already can compute slope / aspect / etc with windows larger than 3x3, and (b) we already have multiple curvature types. From the other half, (c) preprocessing the raster to determine which window size to use in which area, (d) approximating the raster locally using a quadratic surface instead of a plane, and (e) computing geodetic instead of Euclidean distances for rasters that are in lat/lon, the latter two items are straightforward and could be easily added, the remaining (c) is a little unclear in the implementation and in general seems related to performance, instead of trying to figure out how big the computation window should be in different parts of the raster, it might be simpler to just use the biggest possible window size for the entire raster.

I'll check the other two links, too.

Thanks for the post, interesting stuff.

Wonderful, thanks for your interest and summary Adam, this is a nut which would be very useful to crack and I am very interested in any suggestions around how best to achieve this in a meaningful way.

My key interest here is the robust use of LiDAR to derive cohesive slope class groupings for example 0-15 degrees, 15-25 degrees, >25 degrees upon which certain agricultural practices can occur. The slope groupings need to be clean and contiguous, being of a size such that within a farm system, one slope area can be practicably managed differently from another.

The attached is the result of the ESRI tool using a max neighborhood of 100m, separated into the three slope classes above and cleaned up using M9's sieve tool. Though not quite there yet, the resulting slope areas are clean and contiguous and much of the noise has gone.

Though my main interest is slope at the moment, adaptive neighborhood would seem to have meaningful application to many different DEM based metrics.

Attachments:
Slope group example.png

Landsystems Ltd ... Know your land | www.landsystems.co.nz

Looking at an image on the how-surface-parameters-works.htm page, I notice a misleading aspect. The geodetic latitude line should not go through the center of earth. Better image on wikipedia and even more thorough explanation at https://www.e-education.psu.edu/geog862/node/1794